Research On Building Outline Extraction For High Resolution Remote Sensing Image

With the continuous development of remote sensing technology, to obtain high resolution remote sensing images is more fast and easy. Application of high-resolution remote sensing images has become increasingly widespread. To get various types of geographic information by high-resolution remote sensing image is an important application and way for people.This paper mainly studied as follows:(1) I have studied the methods of building outlines extractions with the high resolution remote sensing images which the previous researched, then analyzed and summarized the ways to extract the contour of buildings. The buildings' in high resolution remote sensing images appear as planar structures, the edge features of them are clear, and the color of the roofs is rich and legible, this paper mainly proposed the method to extract the outline of buildings which is based on the images. There are two ways mention in this article, one is the buildings' contour extraction based on edge detection, and the other is based on the color of the images.(2) I have studies the method of building contour extraction of the previous research and divided the building contour extractions into two sections. The first step is the edge detection of building. The second step is contour extraction. In the first step, I have analyzed and summarized the existing edge detection algorithm, then emphasis on improving the Sobel operator. The traditional Sobel operator has two template—horizontal template and vertical template, and in this paper the template Sobel operator using is four directions, and later using eight directions. In the section of the threshold determination, the Otsu's method to calculate the optimal global threshold is used to select an appropriate threshold to judge the values of gradient of the edge. The tests shows that the edge detection is working well, but the result of the contour extraction is not good, and I have analyzed the causes of errors.(3) The high resolution remote sensing images' color information is rich, and there is a big difference in color between the buildings and its' surrounding areas. For this feature, I have to divided the RGB images into three channels. In each channel, the same or similar regions' pixel values are also have the same or similar color or desired average estimate. Based on the desired average estimate, I can do the image segmentation with the high resolution remote sensing images, and then do the edge detection after the completion of the image segmentation. After a series of treatment such as Hough transform, determination of parallel lines, and then the outlines of the building is clear.After a series of simulation tests, it's shown to all: the improved Sobel operator compared with the previous can be more complete response to the edge of the building, and can more effectively avoid the influence caused by the shadow. The building extraction based on color image segmentation can effectively extract the contour of the building, and can effectively reduce the "foreign body with the spectrum, synonyms spectrum" phenomenon.